Hi,

I am fresh to CFD modelling. Could someone please tell me how to judge whether a problem is steady or transient?

As far as I know, for the transient problem, the calculation is performed during each time step until the convergence is achieved, then calculation is moved to the next time step and the programme never stops; for the steady problem, the time step is set to be infinite large, so that the programme stops when it converges.

I have modelled the Ch4 pool fire for both steady and transient problems, and the results are very similar. So what is the point to perform the transient calculation?

Thanks very much for your help

Jianping

Did you have any symmetry BCs in your pool fire?

Thanks, The burner is on the floor, where inflow condition (constant fuel supply) is applied. For the other boundaries, pressure condition are used.

Jianping

If you do what you think is a steady-state problem with the time step set to 0 and it doesn't converge, do you have a numerical problem or is your physical problem really unsteady?

If you do that same problem using a transient analysis and you get convergence to a steady state, that's comforting (not proof but comforting) that your problem really has a steady state. If the problem doesn't have a steady state, you'll hopefully get an approximation to the physical transient flow. And you'll likely learn something very interesting about your flow!

I have another question to ask. For the transient problem, I normally choose time step as 1 second. But it takes more than thousands of iterations to get converged for the first time step. And the strange thing is the results after the first time step more or less agree with the experimental data (when i use the same convergence critirion for both steady and trainsient problems).

I noticed some cfd code either sets the number of iterations for each time step or sets quite large convergent critirion to achieve convergence. Is this sometimes the case to speed up the calculation to reach the phsical time needed?

I think you're asking too general a question.

In my experience, it was not uncommon for more iterations to be needed for a larger time step.

Is a step of one second a large step? Maybe, maybe not. If the physical problem you're trying to simulate has a steady state and you know (experiment?) that any startup transient is much smaller than one second, your simulation might settle on the steady solution - or it might blow up - or it might overshoot and gradually die down to the steady state. I've seen all three.

Whether a time step is 'large' depends on the velocities, transport properties, and mesh sizes as well as expected start up transients.

It's often useful to play with simple problems, things like wave propagation in one space dimension with and without time dependence. You can learn a lot without a lot of computing time, and the results are easy to plot up for study. Good luck!

Yes. As the burner is placed at the centre of the floor, I just model one quarter of the fire with two symmetrical boundaries. so there are actually one inflow boundary, three pressure boundaries and two symmetrical boundaries. Thanks

Many thanks!

Do you mean 1 second as 1 time step is too large. If I chose much smaller time step, I could have achieved convergence with much less iterations? The reason I used 1s is because it is the default value, and I guess I was wrong with this. I will try to use a small time step. I also appreciate your suggestion with some simple cases.

And do you have any suggestion as to how to choose the right convergence criteria. I know some people try to use different Convergence criteria for one case, if there is no big difference between the two, the bigger one is normally adopted for the calculation thereafter. Is this correct?

I really can't give more specific advice because the answer depends so much on your mesh, the velocities, the transport properties (viscosity, etc) and the underlying physical problem. The relationships are complex.

You need to understand the Corant-Friedrichs-Levy (CFL) number, the Cell Reynolds Number, and how they interact with your choice of algorithm.

I'd start (being an old timer) with C. W. Hirt's "Heuristic Stability Theory", published in JCP in the 1960's. I imagine that other readers of this forum can suggest modern text books that cover the choice of algorithm and time step.

Hi Jianping,

This is what I was asking in the first place.

With an unsteady soln, the flame will oscillate back and forth across the plane of symmetry (in reality). As you have set a symmetrical BC, there is no flow through the symmetry plane and therefore this cannot occur and you will get a symmetric steady soln.

I imagine the cause or your 'steady' solution.

Jon